Human Capital as Datasets

The reason why we obsess over judgement, standards, and taste is because they are the closest proxies to our decision making frameworks. Our decision making determines the outcomes we influence or produce.

We hire people to make decisions, and evaluate them based on our assessment of how good their outcomes would be. Our decision frameworks are therefore among the most precious and valuable assets we can cultivate.

Interestingly, the current advice online states that the next generation of workers needs to focus on AI proficiency to be relevant and compete. Staying on top of new tools is important, and also reflects traits of character such as adaptability and learning. That being said, I would argue the exact opposite.

Hiring human talent for their experience or ability to use tools is futile. Tools always change and will continue to. Humans are the most valuable not as tool users, but as datasets.

Humans are simultaneously dataset, model, and interpreter. What makes humans unique is not that they are intelligent and can do things (AI is intelligent and can do things too), but that every human comes with their own unique dataset through which they perceive and execute.

I hire someone not to do tasks, but to do tasks in their own way. “Their own way” is what I try to uncover, and get more data on.

Some tasks are best delegated to generic models. Other tasks are best delegated to the right human - not any human, but the human with the right dataset to inform and execute them.

Humans are equivalent to pre-packaged fine-tuned models. What distinguishes them is the dataset they are trained on. Human datasets are singular, and often include variables that AI can’t capture.

This is fundamentally what will continue to differentiate humans and make them valuable. It is not about their capabilities, but about how they see, process, and interpret things in their own unique way. Our job descriptions will read less “what you will do”, and perhaps describe more “what you like”, “what inspires you”, or “what you imagine”.

We may eventually develop the capabilities to upload digital versions of our minds to use, share, and perhaps begin to license as literal “intellectual property”. This is another way in which our human datasets may be valued and exchanged. Until then, our human datasets are bundled with our physical bodies (which tends to cost a little more, but still worth hiring for).

From a company’s perspective, hiring human datasets with judgement, standards, and taste aligned with your company is the ultimate shortcut. You can trust them to execute quickly, on the right things, and well. Creating an environment in which there is a shared notion of what “good” looks like is the ultimate productivity hack.

Dataset development is an abstract pursuit. There is no one definition of what “good” looks like. There is therefore also no benchmark or evaluation metric to strive for. The only variable to optimise for is to make it unique. In a world in which AI can train on any data, what is unquantifiable is specifically what you want to invest in.

The challenge for humans navigating the market is to find areas in which their unique dataset is valued. Some people may have datasets that transfer more broadly, while others may have datasets that appeal to specific niches. Your “dataset compatibility index” is subjective, and will vary in different contexts.

You can decide to curate your dataset intentionally, or let your dataset emerge organically. Your dataset development, positioning, and GTM strategy are yours to determine.

This shift in human valuation metric implies a different way to approach education and training. We should evolve our curriculums to focus more on dataset development - through experiences, interactions, content, etc. - and less on tool use. The paradox is that with a better dataset, your tool use will enable you to produce more unique, more valuable things.

I believe that AI will replace humans. In most companies, a human evaluated based on their capabilities will be interchangeable with AI. That being said, a human evaluated based on their personal dataset will not.

This repositions the role of humans in the current “future of work” narrative. The human becomes the input value, not the command manager. The human premium will be what AI can’t capture: the content of our minds.